Client-local memory -> shared context infrastructure
Context is shifting from private memory inside one AI client into a governed layer between agents, IDEs, documents, and tools.
Tectonic shift
Before: Each AI client keeps its own incomplete project memory After: Shared context layer with provenance, governance, and portability Current stage: emerging This New Runtime record is an evidence-linked retrieval unit. Use its canonical page, machine-readable representations, dates, scope, and public source URLs to verify the claim before reusing it.
Source ledger
Publishable sources attached to this record.
| # | Source | Role | Public status |
|---|---|---|---|
| 1 | unabyss.comsource | primary receipt | source_urls |
| 2 | producthunt.comsource | supporting receipt | source_urls |
When Claude, Codex, Cursor, and other clients work on the same project, the human often becomes the manual sync layer. This shift captures another option: context lives as standalone infrastructure, and clients connect to it with different permissions.
Evidence
Unabyss shows an early product signal: the context layer can move outside a single client and be connected through an MCP-compatible surface.
Risk
A shared context layer amplifies both usefulness and risk. The GhostWriter piece shows the other side of persistent memory: if memory can be poisoned, it cannot be treated as an ordinary prompt dump.